Impact of Climate and Public Health Interventions on the ... · The association of epidemic growth...

1

Appendix 1 (as supplied by the authors): Protocol

Impact of Climate and Public Health Interventions on the COVID-19 Pandemic:

Prospective Cohort Study

Peter Jüni, Martina Rothenbühler, Pavlos Bobos, Kevin E. Thorpe, Bruno R. da Costa, David N.

Fisman, Arthur S. Slutsky, Dionne Gesink

This supplement contains the following items:

1. Final protocol2. Original protocol3. Summary of changes

Appendix to: Jüni P, Rothenbühler M, Bobos P, et al. Impact of climate and public health interventions on the COVID-19 pandemic: A prospective cohort study. CMAJ 2020. doi: 10.1503/cmaj.200920. Copyright © 2020 Joule Inc. or its licensors

Latitude and COVID-19 epidemic growth - Protocol, Version 1.2, March 25, 2020

1

Study Protocol, Version 1.2

Geographic latitude and epidemic growth during the COVID-19 outbreak

Principal Investigator: Prof. Peter Jüni, Email: [email protected]

Co-Investigators: Prof. Dionne Gesink, Prof. Kevin E. Thorpe, Prof. Arthur S. Slutsky

Background

Seasonality and climate-dependency of influenza are well established. Suggested mechanisms for a slowing down of influenza epidemics in summer months in areas with temperate climate include higher temperature and higher humidity, both potentially decreasing virus survival and transmission, higher ultraviolet radiation, potentially decreasing virus survival, less crowding, and several positive effects of warmer and more humid climate on the host, all potentially decreasing transmission - see (Tamerius et al, 2011) for an overview.

The common factor causally associated with all three major components distinguishing summer and winter months – temperature, humidity and ultraviolet radiation – is latitude (Kharvus and Shelevytsky 2012). Determining latitude is simple and unambiguous and does not require any field work. As we are currently approaching equinox, latitude and its association with temperature, humidity and ultraviolet radiation are particularly easy to interpret.

It is unclear whether COVID-19 is also seasonal and climate-dependent, but many of the suggested explanations for the seasonality of influenza may also be applicable to COVID-19. Evidence for seasonality and climate dependency of COVID-19 would have implications for public health and health care resource planning.

An analysis of absolute rates of reported cases per 1 million population is unlikely to be helpful when determining the association of COVID-19 outbreaks with latitude. Absolute rates are influenced by idiosyncratic processes, such as travel activity, social behavior and timing of arrival of early spreaders who contracted the disease in Wuhan or in the subsequent disease hotspots, South Korea, Iran and Italy, and by alertness, health care resources, and availability and use of tests in individual countries, with potentially dramatic variation in detection and reporting between countries.

An epidemic is typically characterized by an initial stochastic phase followed by exponential growth until the threshold fraction of susceptible individuals is below the inverse of the basic reproduction number R0, or until measures taken start to control the disease outbreak and the threshold fraction of susceptible individuals is below the inverse of the effective reproduction number Re. When either of these points is reached then the growth starts to decelerate.

An analysis of the association of epidemic growth at beginning of the epidemic curve, expressed in relative terms – as a rate ratio comparing the current rate of reported cases with the rate of cases reported a week earlier – in countries and regions currently affected by COVID-19 that



2

are at the beginning of the epidemic curve with latitude, is likely to be more helpful than the analysis of an association of the absolute rate with latitude, as the relative estimates of rate ratios will likely be considerably less affected by between-country variation in detection and reporting of COVID-19 than absolute rates per 1 million population.

Previous study

We performed a previous study according to the pre-specified study protocol version 1.0 dated March 7, 2020, which covered the index period from February 29, 2020 to March 7, 2020. Using data from the online interactive dashboard hosted by the Center for Systems Science and Engineering at Johns Hopkins University, Baltimore and downloaded on March 8, 2020, we calculated the log rate ratio comparing the rate of COVID-19 cases on March 7, 2020 with the rate on February 29, 2020 as a measure of epidemic growth for all geopolitical areas, for which at least 5 COVID-19 cases had been reported by March 7, 2020, excluding China, South Korea, Iran and Italy. We used inverse-variance random-effects regression analysis, univariate and after inclusion of 8 covariates, to regress the log rate ratio against the square of the geographic latitude of analyzed geopolitical areas. Secondary analyses used mean temperature and absolute humidity as independent variables.

Figure. Bubble plot of the rate ratio of COVID-19 on a logarithmic scale on the y-axis against latitude on the x-axis (quadratic term), with prediction lines and 95% confidence interval and the size of the bubbles proportional to the inverse of the variance of log rate ratios.

73 areas with 5569 cases were included. In the univariate analysis, there was a strong association of epidemic growth with latitude (ratio of rate ratios (RRR) per 400 degrees-squared increase, 1.23, 95%-CI 1.14 to 1.23, p


3

0.92, explaining 15.7% of the variation) in univariate analyses, but not independently after inclusion of the 8 covariates in multivariable analyses.

Closer inspection of the data suggested outliers to the top middle and right with implausibly high rate ratios (Figure), which may have resulted in an overestimation of the quadratic relationship between latitude and epidemic growth. These outliers are likely explained by our eligibility criteria of 5 cases accumulated per geopolitical area at the end of the index period of March 7, 2020, which resulted in the inclusion of geopolitical areas that were at the transition from the stochastic phase to exponential growth and may have caused implausible random highs.

We therefore refrained from publishing the results of this study, revised the study protocol without inspection of new data to include more stringent inclusion criteria for geopolitical areas: at least 10 accumulated cases per geopolitical area at the beginning of the index period, and documented local transmission according to the WHO’s Situation Reports.

The initial revision, protocol version 1.1, included a revision of the inclusion criteria, dropped two initially pre-specified covariates for which data was incomplete, added a new covariate, and specified a new index period, March 14 to March 21, 2020 (including vernal equinox), during which the definite study was to be performed.

The current revision, protocol version 1.2, includes an increase of the time lag between exposure and index period to 14 days. With the exposure period set as originally planned from March 7 to March 13, 2020, the new index period was therefore moved to end on March 27, 2020. An additional secondary hypothesis was added on the association of epidemic growth with school closures, restriction of mass gatherings, and social distancing as non-pharmaceutical interventions. Finally, we prespecified the development of two parsimonious models with selection of covariates either based on theoretical grounds or based on a stepwise backward selection.

Proposal

This protocol proposes to determine whether epidemic growth of COVID-19 outbreaks observed during a 7-day index period ending on March 27, 2020, is associated with latitude. An association with latitude would suggest seasonality and climate dependency and could in turn have important implications for health care resource planning and public health strategies.

Primary hypothesis

1. Epidemic growth observed during the index period, as estimated by a rate ratio comparing the event rate at March 27 with the event rate at March 20, 2020, is associated with the square of the latitude in degrees (quadratic relation).

As the index period is near vernal equinox, which will occur on March19, 2020, a simple quadratic relation adequately represents the association of latitude with intensity of insolation and in turn the association of latitude with temperature, humidity and ultraviolet radiation, and therefore with season, without further transformation.



4

Secondary hypotheses

2. Epidemic growth observed during the index period, as estimated by a rate ratio comparing the event rate at March 27 with the event rate at March 20, 2020, is associated with the mean temperature in °C observed during the exposure period (linear relation).

3. Epidemic growth observed during the index period, as estimated by a rate ratio comparing the event rate at March 27 with the event rate at March 20, 2020, is associated with mean absolute humidity observed during the exposure period (linear relation).

4. Epidemic growth observed during the index period, as estimated by a rate ratio comparing the event rate at March 27 with the event rate at March 20, 2020, is associated with the presence or absence of non-pharmaceutical interventions (school closures, restriction of mass gatherings, and measures of social distancing) during the exposure period.

Absolute humidity was more robustly associated with influenza than relative humidity (Tamerius et al, 2011), and is considerably more stable over 24 hours than relative humidity. Both temperature and humidity are likely to be affected by measurement error. Study design and analysis

Primary analysis: Inverse-variance random-effects regression analysis, regressing the log rate ratio of rates of cases on March 27, 2020 divided by rates of cases one week earlier on March 20, 2020 (dependent variable) against the square of the latitude of eligible geopolitical areas (independent variable).

Secondary analyses: Inverse-variance random-effects regression analysis, regressing the log rate ratio of cases in an area (dependent variable) against (1) the mean temperature, and (2) the mean absolute humidity observed during the exposure period from March 7 to March 13, 2020, and (3) school closures, (4) restriction of mass gatherings, and (5) measures of social distancing implemented by mid workweek of the exposure period, Wednesday March 1, 2020 (independent variables).

Analyses addressing primary and secondary hypotheses will be done using a univariate model, and using a multivariable model with simultaneous inclusion of 8 pre-specified covariates (see below); for the analysis of latitude, we will include altitude as an additional covariate. Analyses will be repeated also adjusting for major geographical regions (Asia, Oceania, Europe, Africa, Americas) to account for the geographic progression of the pandemic.

No stepwise selection of covariates will be performed for the multivariable models addressing primary and secondary hypotheses. However, two parsimonious models will be developed in addition: Model 1 with selection of covariates based on theoretical grounds while accounting for clustering and variation of covariates, Model 2 with selection of covariates done using stepwise backward selection based on the adjusted R2 statistic. Model 1 will take precedence over Model



5

2, as Model 2 may be subject to some overfitting. No multiple imputation of covariate data will be done.

To adjust for multiplicity of testing when performing both an unadjusted and an adjusted primary analysis, we will perform a Bonferroni correction and set the alpha level to 0.025 (0.05/2) for both the unadjusted and adjusted analysis. Since this study will be done after finding a highly significant association of epidemic growth with latitude squared (p


6

• Data on the following covariates will be obtained from World Bank and other publicly available sources if necessary and included as covariates in univariate and multivariable regression models: GDP per capita, health expenditure as percent of GDP, life expectancy, percentage of inhabitants aged 65 or over, urban population density.

• We will obtain information about the number of flight passengers per inhabitant from

publicly available sources and included as covariate in univariate and multivariable regression models.

• The distance to the nearest disease hotspot (China, South Korea, Iran or Italy) will be

determined based on coordinates of the country and included as covariate in univariate and multivariable regression models.

• Altitude will be collected for the area’s capital using publicly available sources and included

as covariate in univariate models, and as a covariate in multivariable regression models when analyzing latitude.

• The Infectious Disease Vulnerability Index will be obtained from RAND Corporation and

included as covariate in univariate and multivariable regression models.

March 25, 2020 – Peter Jüni, Dionne Gesink, Kevin E. Thorpe, Arthur S. Slutsky

References

Tamerius James, Nelson Martha I., Zhou Steven Z., Viboud Cécile, Miller Mark A., Alonso Wladimir J. Global Influenza Seasonality: Reconciling Patterns across Temperate and Tropical Regions. Environmental Health Perspectives. 2011 Apr 1;119(4):439–45.

Kharvus V, Shelevytsky I. Geometry and the physics of seasons. Physics Education. 2012;47:680–92.



1

Study Protocol, Version 1.0

Geographic latitude and epidemic growth during the COVID-19 outbreak

Prof. Peter Jüni MD FESC, Applied Health Research Centre (AHRC), Li Ka Shing Knowledge Institute of St. Michael’s Hospital, Department of Medicine and Institute of Health Policy, Management and Evaluation, University of Toronto, Canada. Email: [email protected]

Background

• Seasonality and climate-dependency of influenza are well established. Suggested mechanisms for a slowing down of influenza epidemics in summer months in areas with temperate climate include higher temperature and higher humidity, both potentially decreasing virus survival and transmission, higher ultraviolet radiation, potentially decreasing virus survival, less crowding, and several positive effects of warmer and more humid climate on the host, all potentially decreasing transmission - see (Tamerius et al, 2011) for an overview.

• The common factor causally associated with all three major components distinguishing summer and winter months – temperature, humidity and ultraviolet radiation – is latitude (Kharvus and Shelevytsky 2012). Determining latitude is simple and unambiguous and does not require any field work. As we are currently approaching equinox, latitude and its association with temperature, humidity and ultraviolet radiation are particularly easy to interpret.

• It is unclear whether COVID-19 is also seasonal and climate-dependent, but many of the suggested explanations for the seasonality of influenza may also be applicable to COVID-19. Evidence for seasonality and climate dependency of COVID-19 would have implications for public health and health care resource planning.

• An analysis of absolute rates of reported cases per 1 million population is unlikely to be helpful when determining the association of COVID-19 outbreaks with latitude. Absolute rates are influenced by idiosyncratic processes, such as travel activity, social behavior and timing of arrival of early spreaders who contracted the disease in Wuhan or in the subsequent disease hotspots, South Korea, Iran and Italy, and by alertness, health care resources, and availability and use of tests in individual countries, with potentially dramatic variation in detection and reporting between countries.

• An epidemic is typically characterized by an initial stochastic phase followed by exponential growth until the threshold fraction of susceptible individuals is below the inverse of the basic reproduction number R0, or until measures taken start to control the disease outbreak and the threshold fraction of susceptible individuals is below the inverse of the effective reproduction number Re. When either of these points is reached then the growth starts to decelerate.



2

• An analysis of the association of epidemic growth at beginning of the epidemic curve, expressed in relative terms – as a rate ratio comparing the current rate of reported cases with the rate of cases reported a week earlier – in countries and regions currently affected by COVID-19 that are at the beginning of the epidemic curve with latitude, is likely to be more helpful than the analysis of an association of the absolute rate with latitude, as the relative estimates of rate ratios will likely be considerably less affected by between-country variation in detection and reporting of COVID-19 than absolute rates per 1 million population.

Proposal

• This protocol proposes to determine whether epidemic growth of COVID-19 outbreaks observed during a 7-day index period ending on March 7, 2020, is associated with latitude. An association with latitude would suggest seasonality and climate dependency and could in turn have important implications for health care resource planning and public health strategies.

Primary hypothesis

1. Epidemic growth observed during the index period, as estimated by a rate ratio comparing the event rate at March 7 with the event rate at February 29, 2020, is associated with the square of the latitude in degrees (quadratic relation).

• As the index period is near equinox, which will occur on March 20, a simple quadratic relation is likely to adequately represent the association of latitude with intensity of insolation and in turn the association of latitude with temperature, humidity and ultraviolet radiation, and therefore with season, without further transformation.

Secondary hypotheses

2. Epidemic growth observed during the index period, as estimated by a rate ratio comparing the event rate at March 7 with the event rate at February 29, 2020, is associated with the mean temperature in °C observed during the index period (linear relation).

3. Epidemic growth observed during the index period, as estimated by a rate ratio comparing the event rate at March 7 with the event rate at February 29, 2020, is associated with mean absolute humidity observed during the index period (linear relation).

• Absolute humidity was more robustly associated with influenza than relative humidity (Tamerius et al, 2011), and is considerably more stable over 24 hours than relative humidity.

• Both temperature and humidity are likely to be affected by measurement error given the limited nature of publicly available data on these variables during the index period.



3

Study design and analysis

• Primary analysis: Inverse-variance random-effects regression analysis, regressing the log rate ratio of rates of cases on March 7, 2020 divided by rates of cases one week earlier on February 29, 2020 (dependent variable) against the square of the latitude of areas affected by COVID-19, for which at least 5 cases were reported by March 7, 2020 (independent variable).

• Secondary analyses: Inverse-variance random-effects regression analysis, regressing the log rate ratio of cases in an area (dependent variable) against the average temperature in this area during the index period (independent variable), and regressing the log rate ratio of cases in an area (dependent variable) against mean absolute humidity observed during the index period (independent variable).

• All analyses will be done in a univariate manner, and simultaneously adjusted in a multivariable model for all publicly available covariates (see below) without missing data. No stepwise selection of covariates will be performed. No multiple imputation of covariate data will be done.

• To adjust for multiplicity of testing when performing both an unadjusted and an adjusted primary analysis, we will perform a Bonferroni correction and set the alpha level to 0.025 (0.05/2) for both the unadjusted and adjusted analysis.

• Two sensitivity analyses will be done, (1) restricting the analysis to areas for which at least 10 cases were reported by March 7, 2020, and (2) restricting the analysis to high income countries as defined by the World Bank.

Inclusion criteria

• All areas worldwide for which at least 5 cases were reported by March 7, 2020. Areas will typically be countries, but will be defined as provinces or states in case of countries covering large geographic areas, such as the United States or Canada.

Exclusion criteria

• China will be excluded as China’s epidemic growth has decelerated and the outbreak appears contained.

• South Korea, Italy and Iran will be excluded as their disease outbreak is already fully established, being further ahead on the epidemic curve, with large scale control measures of unclear effectiveness in place.

Data collection

• Latitude of each area will be determined using Google Maps or other sources.

• Temperature and humidity data will be obtained for the area’s capital for the index period using publicly available websites such as weather.com.



4

• Data on the following covariates will be obtained from World Bank and other publicly

available sources if necessary and used for multivariable adjustment of regression models: GDP per capita, health expenditure as percent of GDP, life expectancy, prevalence of undernourishment, GINI index, percentage of inhabitants aged 65 or over, urban population density.

• Whenever possible, we will obtain information about the number of flight passengers per inhabitant from publicly available sources and use as covariate for multivariable adjustment of regression models.

• The distance to the nearest disease hotspot (China, South Korea, Iran or Italy) will be determined based on coordinates of the country and used for multivariable adjustment of regression models.

• Altitude will be collected for the area’s capital using publicly available sources and used for multivariable adjustment of regression models.

March 7, 2020 – Peter Jüni

References

Tamerius James, Nelson Martha I., Zhou Steven Z., Viboud Cécile, Miller Mark A., Alonso Wladimir J. Global Influenza Seasonality: Reconciling Patterns across Temperate and Tropical Regions. Environmental Health Perspectives. 2011 Apr 1;119(4):439–45.

Kharvus V, Shelevytsky I. Geometry and the physics of seasons. Physics Education. 2012;47:680–92.


Protocol: Geographic latitude and epidemic growth during the COVID-19 outbreak Summary of Protocol Changes

The following is a summary of key changes made to Version 1.0 and 1.1 of the protocol. Formatting changes and minor changes (typos, grammatical changes, minor clarifications in wording) are not listed in this table.

Version 1.0 to 1.1 (Date of changes: March 15, 2020): Section Changes

Title Co-investigators, Dionne Gesink, Kevin E. Thorpe and Arthur S. Slutsky, were added.

Previous study A new paragraph describing the previous study conducted for the index period of February 29 to March 7, 2020 and its main results was added.

Proposal The dates of the index period were changed to March 14 to March 21.

Primary hypothesis The dates of the index period were changed.

The wording was changed, indicating that the new index period includes vernal equinox.

Secondary hypothesis The dates of the index period were changed.


The number of pre-specified covariates was specified to be 9.

An explanation was added describing the reason why no further adaptation of alpha was necessary for the definite study conducted for the index period from March 14 to March 21, 2020.

The criteria for the first sensitivity analysis were changed from at least 10 cases reported at the end of the previous index period (March 7, 2020) to at least 20 at the beginning of the current index period (March 14).

Inclusion criteria Inclusion criteria were changed to all geopolitical areas worldwide for which at least 10 cases were reported by March 14, 2020 and for which local transmission was documented according to the WHO’s Situation Report 54.

Data collection An explanation was added describing that temperature and humidity data will be collected for the 7 days prior to the beginning of the index period, March 7 to March 13, 2020, assuming a median incubation period of 5 days, and a median time from symptom onset to diagnosis of 2 days (Lauer et al, 2020).

Prevalence of undernourishment and GINI index were dropped as covariates as their data were found to be incomplete in the previous study.

The Infectious Disease Vulnerability Index was added as a covariate.


Version 1.1 to 1.2 (Date of changes: March 25, 2020):

Previous study A paragraph was added at the end of the section to summarize the changes from protocol version 1.1. to version 1.2.

Proposal The dates of the index period were changed to March 20 to March 27, 2020, corresponding to an increase in the time lag between exposure and index period to 14 days.

Primary hypothesis The dates of the index period were changed to March 20 to March 27, 2020.

Secondary hypothesis The dates of the index period were changed to March 20 to March 27, 2020. An additional secondary hypothesis was specified, indicating that epidemic growth during the index period is associated with the presence or absence of non-pharmaceutical interventions (school closures, restriction of mass gatherings, and measures of social distancing) during the exposure period.


The dates of the index period were changed to March 20 to March 27, 2020. The number of pre-specified covariates was specified to be 8, with altitude used as 9th covariate only for the analysis of the association of epidemic growth with latitude. Secondary analyses were specified for the association of epidemic growth with school closures, restriction of mass gatherings, and measures of social distancing. An additional adjustment for major geographical regions (Asia, Oceania, Europe, Africa, Americas) was introduced to account for the geographic progression of the pandemic. Two parsimonious models were introduced, Model 1 with selection of covariates based on theoretical grounds while accounting for clustering and variation of covariates, Model 2 with selection of covariates done using stepwise backward selection. The date for assessment of the criterion used for the first sensitivity analysis was changed to March 20, 2020.

Inclusion criteria The date for assessment of eligibility criteria was changed to March 20, 2020, and the relevant WHO’s Situation Report for documentation of local transmission to #61.

Data collection The average time lag between transmission and reporting was assumed to be 14 days.


School closures, restriction of mass gatherings, and measures of social distancing were added as secondary covariates.


Study_Protocol_Title_PageStudy_Protocol_v1.2Study_Protocol_v1.0Summary of Protocol Changes_20200325

Impact of Climate and Public Health Interventions on the ... · The association of epidemic growth...

Documents

Transcript of Impact of Climate and Public Health Interventions on the ... · The association of epidemic growth...